datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
   3  * Distributed under the terms of the GNU GPL version 2.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 /* Functions for managing the dp interface/device. */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/init.h>
  14 #include <linux/module.h>
  15 #include <linux/if_arp.h>
  16 #include <linux/if_vlan.h>
  17 #include <linux/in.h>
  18 #include <linux/ip.h>
  19 #include <linux/jhash.h>
  20 #include <linux/delay.h>
  21 #include <linux/time.h>
  22 #include <linux/etherdevice.h>
  23 #include <linux/genetlink.h>
  24 #include <linux/kernel.h>
  25 #include <linux/kthread.h>
  26 #include <linux/mutex.h>
  27 #include <linux/percpu.h>
  28 #include <linux/rcupdate.h>
  29 #include <linux/tcp.h>
  30 #include <linux/udp.h>
  31 #include <linux/version.h>
  32 #include <linux/ethtool.h>
  33 #include <linux/wait.h>
  34 #include <asm/system.h>
  35 #include <asm/div64.h>
  36 #include <asm/bug.h>
  37 #include <linux/highmem.h>
  38 #include <linux/netfilter_bridge.h>
  39 #include <linux/netfilter_ipv4.h>
  40 #include <linux/inetdevice.h>
  41 #include <linux/list.h>
  42 #include <linux/rculist.h>
  43 #include <linux/dmi.h>
  44 #include <net/inet_ecn.h>
  45 #include <net/genetlink.h>
  46
  47 #include "openvswitch/datapath-protocol.h"
  48 #include "checksum.h"
  49 #include "datapath.h"
  50 #include "actions.h"
  51 #include "flow.h"
  52 #include "table.h"
  53 #include "vlan.h"
  54 #include "vport-internal_dev.h"
  55
  56 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  57     LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
  58 #error Kernels before 2.6.18 or after 3.0 are not supported by this version of Open vSwitch.
  59 #endif
  60
  61 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  62 EXPORT_SYMBOL(dp_ioctl_hook);
  63
  64 /**
  65  * DOC: Locking:
  66  *
  67  * Writes to device state (add/remove datapath, port, set operations on vports,
  68  * etc.) are protected by RTNL.
  69  *
  70  * Writes to other state (flow table modifications, set miscellaneous datapath
  71  * parameters such as drop frags, etc.) are protected by genl_mutex.  The RTNL
  72  * lock nests inside genl_mutex.
  73  *
  74  * Reads are protected by RCU.
  75  *
  76  * There are a few special cases (mostly stats) that have their own
  77  * synchronization but they nest under all of above and don't interact with
  78  * each other.
  79  */
  80
  81 /* Global list of datapaths to enable dumping them all out.
  82  * Protected by genl_mutex.
  83  */
  84 static LIST_HEAD(dps);
  85
  86 static struct vport *new_vport(const struct vport_parms *);
  87 static int queue_userspace_packets(struct datapath *, struct sk_buff *,
  88                                  const struct dp_upcall_info *);
  89
  90 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  91 struct datapath *get_dp(int dp_ifindex)
  92 {
  93         struct datapath *dp = NULL;
  94         struct net_device *dev;
  95
  96         rcu_read_lock();
  97         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
  98         if (dev) {
  99                 struct vport *vport = internal_dev_get_vport(dev);
 100                 if (vport)
 101                         dp = vport->dp;
 102         }
 103         rcu_read_unlock();
 104
 105         return dp;
 106 }
 107 EXPORT_SYMBOL_GPL(get_dp);
 108
 109 /* Must be called with genl_mutex. */
 110 static struct tbl *get_table_protected(struct datapath *dp)
 111 {
 112         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
 113 }
 114
 115 /* Must be called with rcu_read_lock or RTNL lock. */
 116 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 117 {
 118         return rcu_dereference_rtnl(dp->ports[port_no]);
 119 }
 120
 121 /* Must be called with rcu_read_lock or RTNL lock. */
 122 const char *dp_name(const struct datapath *dp)
 123 {
 124         return vport_get_name(rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]));
 125 }
 126
 127 static inline size_t br_nlmsg_size(void)
 128 {
 129         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 130                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 131                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 132                + nla_total_size(4) /* IFLA_MASTER */
 133                + nla_total_size(4) /* IFLA_MTU */
 134                + nla_total_size(4) /* IFLA_LINK */
 135                + nla_total_size(1); /* IFLA_OPERSTATE */
 136 }
 137
 138 /* Caller must hold RTNL lock. */
 139 static int dp_fill_ifinfo(struct sk_buff *skb,
 140                           const struct vport *port,
 141                           int event, unsigned int flags)
 142 {
 143         struct datapath *dp = port->dp;
 144         int ifindex = vport_get_ifindex(port);
 145         struct ifinfomsg *hdr;
 146         struct nlmsghdr *nlh;
 147
 148         if (ifindex < 0)
 149                 return ifindex;
 150
 151         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 152         if (nlh == NULL)
 153                 return -EMSGSIZE;
 154
 155         hdr = nlmsg_data(nlh);
 156         hdr->ifi_family = AF_BRIDGE;
 157         hdr->__ifi_pad = 0;
 158         hdr->ifi_type = ARPHRD_ETHER;
 159         hdr->ifi_index = ifindex;
 160         hdr->ifi_flags = vport_get_flags(port);
 161         hdr->ifi_change = 0;
 162
 163         NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
 164         NLA_PUT_U32(skb, IFLA_MASTER,
 165                 vport_get_ifindex(get_vport_protected(dp, OVSP_LOCAL)));
 166         NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 167 #ifdef IFLA_OPERSTATE
 168         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 169                    vport_is_running(port)
 170                         ? vport_get_operstate(port)
 171                         : IF_OPER_DOWN);
 172 #endif
 173
 174         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
 175
 176         return nlmsg_end(skb, nlh);
 177
 178 nla_put_failure:
 179         nlmsg_cancel(skb, nlh);
 180         return -EMSGSIZE;
 181 }
 182
 183 /* Caller must hold RTNL lock. */
 184 static void dp_ifinfo_notify(int event, struct vport *port)
 185 {
 186         struct sk_buff *skb;
 187         int err = -ENOBUFS;
 188
 189         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 190         if (skb == NULL)
 191                 goto errout;
 192
 193         err = dp_fill_ifinfo(skb, port, event, 0);
 194         if (err < 0) {
 195                 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 196                 WARN_ON(err == -EMSGSIZE);
 197                 kfree_skb(skb);
 198                 goto errout;
 199         }
 200         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 201         return;
 202 errout:
 203         if (err < 0)
 204                 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 205 }
 206
 207 static void release_dp(struct kobject *kobj)
 208 {
 209         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 210         kfree(dp);
 211 }
 212
 213 static struct kobj_type dp_ktype = {
 214         .release = release_dp
 215 };
 216
 217 static void destroy_dp_rcu(struct rcu_head *rcu)
 218 {
 219         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 220
 221         tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
 222         free_percpu(dp->stats_percpu);
 223         kobject_put(&dp->ifobj);
 224 }
 225
 226 /* Called with RTNL lock and genl_lock. */
 227 static struct vport *new_vport(const struct vport_parms *parms)
 228 {
 229         struct vport *vport;
 230
 231         vport = vport_add(parms);
 232         if (!IS_ERR(vport)) {
 233                 struct datapath *dp = parms->dp;
 234
 235                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 236                 list_add(&vport->node, &dp->port_list);
 237
 238                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 239         }
 240
 241         return vport;
 242 }
 243
 244 /* Called with RTNL lock. */
 245 int dp_detach_port(struct vport *p)
 246 {
 247         ASSERT_RTNL();
 248
 249         if (p->port_no != OVSP_LOCAL)
 250                 dp_sysfs_del_if(p);
 251         dp_ifinfo_notify(RTM_DELLINK, p);
 252
 253         /* First drop references to device. */
 254         list_del(&p->node);
 255         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 256
 257         /* Then destroy it. */
 258         return vport_del(p);
 259 }
 260
 261 /* Must be called with rcu_read_lock. */
 262 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 263 {
 264         struct datapath *dp = p->dp;
 265         struct dp_stats_percpu *stats;
 266         int stats_counter_off;
 267         int error;
 268
 269         OVS_CB(skb)->vport = p;
 270
 271         if (!OVS_CB(skb)->flow) {
 272                 struct sw_flow_key key;
 273                 struct tbl_node *flow_node;
 274                 int key_len;
 275                 bool is_frag;
 276
 277                 /* Extract flow from 'skb' into 'key'. */
 278                 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
 279                 if (unlikely(error)) {
 280                         kfree_skb(skb);
 281                         return;
 282                 }
 283
 284                 if (is_frag && dp->drop_frags) {
 285                         consume_skb(skb);
 286                         stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
 287                         goto out;
 288                 }
 289
 290                 /* Look up flow. */
 291                 flow_node = tbl_lookup(rcu_dereference(dp->table), &key, key_len,
 292                                        flow_hash(&key, key_len), flow_cmp);
 293                 if (unlikely(!flow_node)) {
 294                         struct dp_upcall_info upcall;
 295
 296                         upcall.cmd = OVS_PACKET_CMD_MISS;
 297                         upcall.key = &key;
 298                         upcall.userdata = 0;
 299                         upcall.sample_pool = 0;
 300                         upcall.actions = NULL;
 301                         upcall.actions_len = 0;
 302                         dp_upcall(dp, skb, &upcall);
 303                         stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
 304                         goto out;
 305                 }
 306
 307                 OVS_CB(skb)->flow = flow_cast(flow_node);
 308         }
 309
 310         stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
 311         flow_used(OVS_CB(skb)->flow, skb);
 312         execute_actions(dp, skb);
 313
 314 out:
 315         /* Update datapath statistics. */
 316         local_bh_disable();
 317         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 318
 319         write_seqcount_begin(&stats->seqlock);
 320         (*(u64 *)((u8 *)stats + stats_counter_off))++;
 321         write_seqcount_end(&stats->seqlock);
 322
 323         local_bh_enable();
 324 }
 325
 326 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 327 {
 328         u16 csum_start, csum_offset;
 329         __wsum csum;
 330
 331         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 332         csum_start -= skb_headroom(skb);
 333
 334         skb_copy_bits(skb, 0, to, csum_start);
 335
 336         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 337                                       skb->len - csum_start, 0);
 338         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 339 }
 340
 341 static struct genl_family dp_packet_genl_family = {
 342         .id = GENL_ID_GENERATE,
 343         .hdrsize = sizeof(struct ovs_header),
 344         .name = OVS_PACKET_FAMILY,
 345         .version = 1,
 346         .maxattr = OVS_PACKET_ATTR_MAX
 347 };
 348
 349 /* Generic Netlink multicast groups for upcalls.
 350  *
 351  * We really want three unique multicast groups per datapath, but we can't even
 352  * get one, because genl_register_mc_group() takes genl_lock, which is also
 353  * held during Generic Netlink message processing, so trying to acquire
 354  * multicast groups during OVS_DP_NEW processing deadlocks.  Instead, we
 355  * preallocate a few groups and use them round-robin for datapaths.  Collision
 356  * isn't fatal--multicast listeners should check that the family is the one
 357  * that they want and discard others--but it wastes time and memory to receive
 358  * unwanted messages.
 359  */
 360 #define PACKET_N_MC_GROUPS 16
 361 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
 362
 363 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
 364 {
 365         u32 idx;
 366         BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
 367
 368         idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
 369         return packet_mc_groups[idx].id;
 370 }
 371
 372 static int packet_register_mc_groups(void)
 373 {
 374         int i;
 375
 376         for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
 377                 struct genl_multicast_group *group = &packet_mc_groups[i];
 378                 int error;
 379
 380                 sprintf(group->name, "packet%d", i);
 381                 error = genl_register_mc_group(&dp_packet_genl_family, group);
 382                 if (error)
 383                         return error;
 384         }
 385         return 0;
 386 }
 387
 388 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
 389 {
 390         struct dp_stats_percpu *stats;
 391         int err;
 392
 393         WARN_ON_ONCE(skb_shared(skb));
 394
 395         forward_ip_summed(skb, true);
 396
 397         /* Break apart GSO packets into their component pieces.  Otherwise
 398          * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
 399         if (skb_is_gso(skb)) {
 400                 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 401
 402                 if (IS_ERR(nskb)) {
 403                         kfree_skb(skb);
 404                         err = PTR_ERR(nskb);
 405                         goto err;
 406                 }
 407                 consume_skb(skb);
 408                 skb = nskb;
 409         }
 410
 411         err = queue_userspace_packets(dp, skb, upcall_info);
 412         if (err)
 413                 goto err;
 414
 415         return 0;
 416
 417 err:
 418         local_bh_disable();
 419         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 420
 421         write_seqcount_begin(&stats->seqlock);
 422         stats->n_lost++;
 423         write_seqcount_end(&stats->seqlock);
 424
 425         local_bh_enable();
 426
 427         return err;
 428 }
 429
 430 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
 431  * 'upcall_info'.  There will be only one packet unless we broke up a GSO
 432  * packet.
 433  */
 434 static int queue_userspace_packets(struct datapath *dp, struct sk_buff *skb,
 435                                  const struct dp_upcall_info *upcall_info)
 436 {
 437         u32 group = packet_mc_group(dp, upcall_info->cmd);
 438         struct sk_buff *nskb;
 439         int err;
 440
 441         do {
 442                 struct ovs_header *upcall;
 443                 struct sk_buff *user_skb; /* to be queued to userspace */
 444                 struct nlattr *nla;
 445                 unsigned int len;
 446
 447                 nskb = skb->next;
 448                 skb->next = NULL;
 449
 450                 err = vlan_deaccel_tag(skb);
 451                 if (unlikely(err))
 452                         goto err_kfree_skbs;
 453
 454                 if (nla_attr_size(skb->len) > USHRT_MAX)
 455                         goto err_kfree_skbs;
 456
 457                 len = sizeof(struct ovs_header);
 458                 len += nla_total_size(skb->len);
 459                 len += nla_total_size(FLOW_BUFSIZE);
 460                 if (upcall_info->userdata)
 461                         len += nla_total_size(8);
 462                 if (upcall_info->sample_pool)
 463                         len += nla_total_size(4);
 464                 if (upcall_info->actions_len)
 465                         len += nla_total_size(upcall_info->actions_len);
 466
 467                 user_skb = genlmsg_new(len, GFP_ATOMIC);
 468                 if (!user_skb) {
 469                         netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
 470                         goto err_kfree_skbs;
 471                 }
 472
 473                 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
 474                 upcall->dp_ifindex = dp->dp_ifindex;
 475
 476                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 477                 flow_to_nlattrs(upcall_info->key, user_skb);
 478                 nla_nest_end(user_skb, nla);
 479
 480                 if (upcall_info->userdata)
 481                         nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA, upcall_info->userdata);
 482                 if (upcall_info->sample_pool)
 483                         nla_put_u32(user_skb, OVS_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
 484                 if (upcall_info->actions_len) {
 485                         const struct nlattr *actions = upcall_info->actions;
 486                         u32 actions_len = upcall_info->actions_len;
 487
 488                         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 489                         memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
 490                         nla_nest_end(user_skb, nla);
 491                 }
 492
 493                 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
 494                 if (skb->ip_summed == CHECKSUM_PARTIAL)
 495                         copy_and_csum_skb(skb, nla_data(nla));
 496                 else
 497                         skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 498
 499                 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
 500                 if (err)
 501                         goto err_kfree_skbs;
 502
 503                 consume_skb(skb);
 504                 skb = nskb;
 505         } while (skb);
 506         return 0;
 507
 508 err_kfree_skbs:
 509         kfree_skb(skb);
 510         while ((skb = nskb) != NULL) {
 511                 nskb = skb->next;
 512                 kfree_skb(skb);
 513         }
 514         return err;
 515 }
 516
 517 /* Called with genl_mutex. */
 518 static int flush_flows(int dp_ifindex)
 519 {
 520         struct tbl *old_table;
 521         struct tbl *new_table;
 522         struct datapath *dp;
 523
 524         dp = get_dp(dp_ifindex);
 525         if (!dp)
 526                 return -ENODEV;
 527
 528         old_table = get_table_protected(dp);
 529         new_table = tbl_create(TBL_MIN_BUCKETS);
 530         if (!new_table)
 531                 return -ENOMEM;
 532
 533         rcu_assign_pointer(dp->table, new_table);
 534
 535         tbl_deferred_destroy(old_table, flow_free_tbl);
 536
 537         return 0;
 538 }
 539
 540 static int validate_actions(const struct nlattr *attr)
 541 {
 542         const struct nlattr *a;
 543         int rem;
 544
 545         nla_for_each_nested(a, attr, rem) {
 546                 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
 547                         [OVS_ACTION_ATTR_OUTPUT] = 4,
 548                         [OVS_ACTION_ATTR_USERSPACE] = 8,
 549                         [OVS_ACTION_ATTR_SET_DL_TCI] = 2,
 550                         [OVS_ACTION_ATTR_STRIP_VLAN] = 0,
 551                         [OVS_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
 552                         [OVS_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
 553                         [OVS_ACTION_ATTR_SET_NW_SRC] = 4,
 554                         [OVS_ACTION_ATTR_SET_NW_DST] = 4,
 555                         [OVS_ACTION_ATTR_SET_NW_TOS] = 1,
 556                         [OVS_ACTION_ATTR_SET_TP_SRC] = 2,
 557                         [OVS_ACTION_ATTR_SET_TP_DST] = 2,
 558                         [OVS_ACTION_ATTR_SET_TUNNEL] = 8,
 559                         [OVS_ACTION_ATTR_SET_PRIORITY] = 4,
 560                         [OVS_ACTION_ATTR_POP_PRIORITY] = 0,
 561                 };
 562                 int type = nla_type(a);
 563
 564                 if (type > OVS_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
 565                         return -EINVAL;
 566
 567                 switch (type) {
 568                 case OVS_ACTION_ATTR_UNSPEC:
 569                         return -EINVAL;
 570
 571                 case OVS_ACTION_ATTR_USERSPACE:
 572                 case OVS_ACTION_ATTR_STRIP_VLAN:
 573                 case OVS_ACTION_ATTR_SET_DL_SRC:
 574                 case OVS_ACTION_ATTR_SET_DL_DST:
 575                 case OVS_ACTION_ATTR_SET_NW_SRC:
 576                 case OVS_ACTION_ATTR_SET_NW_DST:
 577                 case OVS_ACTION_ATTR_SET_TP_SRC:
 578                 case OVS_ACTION_ATTR_SET_TP_DST:
 579                 case OVS_ACTION_ATTR_SET_TUNNEL:
 580                 case OVS_ACTION_ATTR_SET_PRIORITY:
 581                 case OVS_ACTION_ATTR_POP_PRIORITY:
 582                         /* No validation needed. */
 583                         break;
 584
 585                 case OVS_ACTION_ATTR_OUTPUT:
 586                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 587                                 return -EINVAL;
 588                         break;
 589
 590                 case OVS_ACTION_ATTR_SET_DL_TCI:
 591                         if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
 592                                 return -EINVAL;
 593                         break;
 594
 595                 case OVS_ACTION_ATTR_SET_NW_TOS:
 596                         if (nla_get_u8(a) & INET_ECN_MASK)
 597                                 return -EINVAL;
 598                         break;
 599
 600                 default:
 601                         return -EOPNOTSUPP;
 602                 }
 603         }
 604
 605         if (rem > 0)
 606                 return -EINVAL;
 607
 608         return 0;
 609 }
 610 static void clear_stats(struct sw_flow *flow)
 611 {
 612         flow->used = 0;
 613         flow->tcp_flags = 0;
 614         flow->packet_count = 0;
 615         flow->byte_count = 0;
 616 }
 617
 618 /* Called with genl_mutex. */
 619 static int expand_table(struct datapath *dp)
 620 {
 621         struct tbl *old_table = get_table_protected(dp);
 622         struct tbl *new_table;
 623
 624         new_table = tbl_expand(old_table);
 625         if (IS_ERR(new_table)) {
 626                 if (PTR_ERR(new_table) != -ENOSPC)
 627                         return PTR_ERR(new_table);
 628         } else {
 629                 rcu_assign_pointer(dp->table, new_table);
 630                 tbl_deferred_destroy(old_table, NULL);
 631         }
 632
 633         return 0;
 634 }
 635
 636 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 637 {
 638         struct ovs_header *ovs_header = info->userhdr;
 639         struct nlattr **a = info->attrs;
 640         struct sw_flow_actions *acts;
 641         struct sk_buff *packet;
 642         struct sw_flow *flow;
 643         struct datapath *dp;
 644         struct ethhdr *eth;
 645         bool is_frag;
 646         int len;
 647         int err;
 648         int key_len;
 649
 650         err = -EINVAL;
 651         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 652             !a[OVS_PACKET_ATTR_ACTIONS] ||
 653             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 654                 goto err;
 655
 656         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS]);
 657         if (err)
 658                 goto err;
 659
 660         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 661         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 662         err = -ENOMEM;
 663         if (!packet)
 664                 goto err;
 665         skb_reserve(packet, NET_IP_ALIGN);
 666
 667         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 668
 669         skb_reset_mac_header(packet);
 670         eth = eth_hdr(packet);
 671
 672         /* Normally, setting the skb 'protocol' field would be handled by a
 673          * call to eth_type_trans(), but it assumes there's a sending
 674          * device, which we may not have. */
 675         if (ntohs(eth->h_proto) >= 1536)
 676                 packet->protocol = eth->h_proto;
 677         else
 678                 packet->protocol = htons(ETH_P_802_2);
 679
 680         /* Build an sw_flow for sending this packet. */
 681         flow = flow_alloc();
 682         err = PTR_ERR(flow);
 683         if (IS_ERR(flow))
 684                 goto err_kfree_skb;
 685
 686         err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
 687         if (err)
 688                 goto err_flow_put;
 689         flow->tbl_node.hash = flow_hash(&flow->key, key_len);
 690
 691         err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
 692                                          &flow->key.eth.tun_id,
 693                                          a[OVS_PACKET_ATTR_KEY]);
 694         if (err)
 695                 goto err_flow_put;
 696
 697         acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 698         err = PTR_ERR(acts);
 699         if (IS_ERR(acts))
 700                 goto err_flow_put;
 701         rcu_assign_pointer(flow->sf_acts, acts);
 702
 703         OVS_CB(packet)->flow = flow;
 704
 705         rcu_read_lock();
 706         dp = get_dp(ovs_header->dp_ifindex);
 707         err = -ENODEV;
 708         if (!dp)
 709                 goto err_unlock;
 710         err = execute_actions(dp, packet);
 711         rcu_read_unlock();
 712
 713         flow_put(flow);
 714         return err;
 715
 716 err_unlock:
 717         rcu_read_unlock();
 718 err_flow_put:
 719         flow_put(flow);
 720 err_kfree_skb:
 721         kfree_skb(packet);
 722 err:
 723         return err;
 724 }
 725
 726 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 727         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 728         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 729         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 730 };
 731
 732 static struct genl_ops dp_packet_genl_ops[] = {
 733         { .cmd = OVS_PACKET_CMD_EXECUTE,
 734           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 735           .policy = packet_policy,
 736           .doit = ovs_packet_cmd_execute
 737         }
 738 };
 739
 740 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 741 {
 742         int i;
 743         struct tbl *table = get_table_protected(dp);
 744
 745         stats->n_flows = tbl_count(table);
 746
 747         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 748         for_each_possible_cpu(i) {
 749                 const struct dp_stats_percpu *percpu_stats;
 750                 struct dp_stats_percpu local_stats;
 751                 unsigned seqcount;
 752
 753                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 754
 755                 do {
 756                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 757                         local_stats = *percpu_stats;
 758                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 759
 760                 stats->n_frags += local_stats.n_frags;
 761                 stats->n_hit += local_stats.n_hit;
 762                 stats->n_missed += local_stats.n_missed;
 763                 stats->n_lost += local_stats.n_lost;
 764         }
 765 }
 766
 767 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
 768  * Called with RTNL lock.
 769  */
 770 int dp_min_mtu(const struct datapath *dp)
 771 {
 772         struct vport *p;
 773         int mtu = 0;
 774
 775         ASSERT_RTNL();
 776
 777         list_for_each_entry (p, &dp->port_list, node) {
 778                 int dev_mtu;
 779
 780                 /* Skip any internal ports, since that's what we're trying to
 781                  * set. */
 782                 if (is_internal_vport(p))
 783                         continue;
 784
 785                 dev_mtu = vport_get_mtu(p);
 786                 if (!dev_mtu)
 787                         continue;
 788                 if (!mtu || dev_mtu < mtu)
 789                         mtu = dev_mtu;
 790         }
 791
 792         return mtu ? mtu : ETH_DATA_LEN;
 793 }
 794
 795 /* Sets the MTU of all datapath devices to the minimum of the ports
 796  * Called with RTNL lock.
 797  */
 798 void set_internal_devs_mtu(const struct datapath *dp)
 799 {
 800         struct vport *p;
 801         int mtu;
 802
 803         ASSERT_RTNL();
 804
 805         mtu = dp_min_mtu(dp);
 806
 807         list_for_each_entry (p, &dp->port_list, node) {
 808                 if (is_internal_vport(p))
 809                         vport_set_mtu(p, mtu);
 810         }
 811 }
 812
 813 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 814         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 815         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 816         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 817 };
 818
 819 static struct genl_family dp_flow_genl_family = {
 820         .id = GENL_ID_GENERATE,
 821         .hdrsize = sizeof(struct ovs_header),
 822         .name = OVS_FLOW_FAMILY,
 823         .version = 1,
 824         .maxattr = OVS_FLOW_ATTR_MAX
 825 };
 826
 827 static struct genl_multicast_group dp_flow_multicast_group = {
 828         .name = OVS_FLOW_MCGROUP
 829 };
 830
 831 /* Called with genl_lock. */
 832 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 833                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 834 {
 835         const int skb_orig_len = skb->len;
 836         const struct sw_flow_actions *sf_acts;
 837         struct ovs_flow_stats stats;
 838         struct ovs_header *ovs_header;
 839         struct nlattr *nla;
 840         unsigned long used;
 841         u8 tcp_flags;
 842         int err;
 843
 844         sf_acts = rcu_dereference_protected(flow->sf_acts,
 845                                             lockdep_genl_is_held());
 846
 847         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 848         if (!ovs_header)
 849                 return -EMSGSIZE;
 850
 851         ovs_header->dp_ifindex = dp->dp_ifindex;
 852
 853         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 854         if (!nla)
 855                 goto nla_put_failure;
 856         err = flow_to_nlattrs(&flow->key, skb);
 857         if (err)
 858                 goto error;
 859         nla_nest_end(skb, nla);
 860
 861         spin_lock_bh(&flow->lock);
 862         used = flow->used;
 863         stats.n_packets = flow->packet_count;
 864         stats.n_bytes = flow->byte_count;
 865         tcp_flags = flow->tcp_flags;
 866         spin_unlock_bh(&flow->lock);
 867
 868         if (used)
 869                 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
 870
 871         if (stats.n_packets)
 872                 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
 873
 874         if (tcp_flags)
 875                 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 876
 877         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 878          * this is the first flow to be dumped into 'skb'.  This is unusual for
 879          * Netlink but individual action lists can be longer than
 880          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 881          * The userspace caller can always fetch the actions separately if it
 882          * really wants them.  (Most userspace callers in fact don't care.)
 883          *
 884          * This can only fail for dump operations because the skb is always
 885          * properly sized for single flows.
 886          */
 887         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 888                       sf_acts->actions);
 889         if (err < 0 && skb_orig_len)
 890                 goto error;
 891
 892         return genlmsg_end(skb, ovs_header);
 893
 894 nla_put_failure:
 895         err = -EMSGSIZE;
 896 error:
 897         genlmsg_cancel(skb, ovs_header);
 898         return err;
 899 }
 900
 901 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 902 {
 903         const struct sw_flow_actions *sf_acts;
 904         int len;
 905
 906         sf_acts = rcu_dereference_protected(flow->sf_acts,
 907                                             lockdep_genl_is_held());
 908
 909         len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
 910         len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 911         len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
 912         len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
 913         len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
 914         return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
 915 }
 916
 917 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 918                                                u32 pid, u32 seq, u8 cmd)
 919 {
 920         struct sk_buff *skb;
 921         int retval;
 922
 923         skb = ovs_flow_cmd_alloc_info(flow);
 924         if (!skb)
 925                 return ERR_PTR(-ENOMEM);
 926
 927         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 928         BUG_ON(retval < 0);
 929         return skb;
 930 }
 931
 932 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 933 {
 934         struct nlattr **a = info->attrs;
 935         struct ovs_header *ovs_header = info->userhdr;
 936         struct tbl_node *flow_node;
 937         struct sw_flow_key key;
 938         struct sw_flow *flow;
 939         struct sk_buff *reply;
 940         struct datapath *dp;
 941         struct tbl *table;
 942         u32 hash;
 943         int error;
 944         int key_len;
 945
 946         /* Extract key. */
 947         error = -EINVAL;
 948         if (!a[OVS_FLOW_ATTR_KEY])
 949                 goto error;
 950         error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 951         if (error)
 952                 goto error;
 953
 954         /* Validate actions. */
 955         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 956                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS]);
 957                 if (error)
 958                         goto error;
 959         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 960                 error = -EINVAL;
 961                 goto error;
 962         }
 963
 964         dp = get_dp(ovs_header->dp_ifindex);
 965         error = -ENODEV;
 966         if (!dp)
 967                 goto error;
 968
 969         hash = flow_hash(&key, key_len);
 970         table = get_table_protected(dp);
 971         flow_node = tbl_lookup(table, &key, key_len, hash, flow_cmp);
 972         if (!flow_node) {
 973                 struct sw_flow_actions *acts;
 974
 975                 /* Bail out if we're not allowed to create a new flow. */
 976                 error = -ENOENT;
 977                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 978                         goto error;
 979
 980                 /* Expand table, if necessary, to make room. */
 981                 if (tbl_count(table) >= tbl_n_buckets(table)) {
 982                         error = expand_table(dp);
 983                         if (error)
 984                                 goto error;
 985                         table = get_table_protected(dp);
 986                 }
 987
 988                 /* Allocate flow. */
 989                 flow = flow_alloc();
 990                 if (IS_ERR(flow)) {
 991                         error = PTR_ERR(flow);
 992                         goto error;
 993                 }
 994                 flow->key = key;
 995                 clear_stats(flow);
 996
 997                 /* Obtain actions. */
 998                 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
 999                 error = PTR_ERR(acts);
1000                 if (IS_ERR(acts))
1001                         goto error_free_flow;
1002                 rcu_assign_pointer(flow->sf_acts, acts);
1003
1004                 /* Put flow in bucket. */
1005                 error = tbl_insert(table, &flow->tbl_node, hash);
1006                 if (error)
1007                         goto error_free_flow;
1008
1009                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1010                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1011         } else {
1012                 /* We found a matching flow. */
1013                 struct sw_flow_actions *old_acts;
1014
1015                 /* Bail out if we're not allowed to modify an existing flow.
1016                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1017                  * because Generic Netlink treats the latter as a dump
1018                  * request.  We also accept NLM_F_EXCL in case that bug ever
1019                  * gets fixed.
1020                  */
1021                 error = -EEXIST;
1022                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1023                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1024                         goto error;
1025
1026                 /* Update actions. */
1027                 flow = flow_cast(flow_node);
1028                 old_acts = rcu_dereference_protected(flow->sf_acts,
1029                                                      lockdep_genl_is_held());
1030                 if (a[OVS_FLOW_ATTR_ACTIONS] &&
1031                     (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
1032                      memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
1033                             old_acts->actions_len))) {
1034                         struct sw_flow_actions *new_acts;
1035
1036                         new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1037                         error = PTR_ERR(new_acts);
1038                         if (IS_ERR(new_acts))
1039                                 goto error;
1040
1041                         rcu_assign_pointer(flow->sf_acts, new_acts);
1042                         flow_deferred_free_acts(old_acts);
1043                 }
1044
1045                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1046                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1047
1048                 /* Clear stats. */
1049                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1050                         spin_lock_bh(&flow->lock);
1051                         clear_stats(flow);
1052                         spin_unlock_bh(&flow->lock);
1053                 }
1054         }
1055
1056         if (!IS_ERR(reply))
1057                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1058                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1059         else
1060                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1061                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1062         return 0;
1063
1064 error_free_flow:
1065         flow_put(flow);
1066 error:
1067         return error;
1068 }
1069
1070 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1071 {
1072         struct nlattr **a = info->attrs;
1073         struct ovs_header *ovs_header = info->userhdr;
1074         struct sw_flow_key key;
1075         struct tbl_node *flow_node;
1076         struct sk_buff *reply;
1077         struct sw_flow *flow;
1078         struct datapath *dp;
1079         struct tbl *table;
1080         int err;
1081         int key_len;
1082
1083         if (!a[OVS_FLOW_ATTR_KEY])
1084                 return -EINVAL;
1085         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1086         if (err)
1087                 return err;
1088
1089         dp = get_dp(ovs_header->dp_ifindex);
1090         if (!dp)
1091                 return -ENODEV;
1092
1093         table = get_table_protected(dp);
1094         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1095                                flow_cmp);
1096         if (!flow_node)
1097                 return -ENOENT;
1098
1099         flow = flow_cast(flow_node);
1100         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
1101         if (IS_ERR(reply))
1102                 return PTR_ERR(reply);
1103
1104         return genlmsg_reply(reply, info);
1105 }
1106
1107 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1108 {
1109         struct nlattr **a = info->attrs;
1110         struct ovs_header *ovs_header = info->userhdr;
1111         struct sw_flow_key key;
1112         struct tbl_node *flow_node;
1113         struct sk_buff *reply;
1114         struct sw_flow *flow;
1115         struct datapath *dp;
1116         struct tbl *table;
1117         int err;
1118         int key_len;
1119
1120         if (!a[OVS_FLOW_ATTR_KEY])
1121                 return flush_flows(ovs_header->dp_ifindex);
1122         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1123         if (err)
1124                 return err;
1125
1126         dp = get_dp(ovs_header->dp_ifindex);
1127         if (!dp)
1128                 return -ENODEV;
1129
1130         table = get_table_protected(dp);
1131         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1132                                flow_cmp);
1133         if (!flow_node)
1134                 return -ENOENT;
1135         flow = flow_cast(flow_node);
1136
1137         reply = ovs_flow_cmd_alloc_info(flow);
1138         if (!reply)
1139                 return -ENOMEM;
1140
1141         err = tbl_remove(table, flow_node);
1142         if (err) {
1143                 kfree_skb(reply);
1144                 return err;
1145         }
1146
1147         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1148                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1149         BUG_ON(err < 0);
1150
1151         flow_deferred_free(flow);
1152
1153         genl_notify(reply, genl_info_net(info), info->snd_pid,
1154                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1155         return 0;
1156 }
1157
1158 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1159 {
1160         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1161         struct datapath *dp;
1162
1163         dp = get_dp(ovs_header->dp_ifindex);
1164         if (!dp)
1165                 return -ENODEV;
1166
1167         for (;;) {
1168                 struct tbl_node *flow_node;
1169                 struct sw_flow *flow;
1170                 u32 bucket, obj;
1171
1172                 bucket = cb->args[0];
1173                 obj = cb->args[1];
1174                 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1175                 if (!flow_node)
1176                         break;
1177
1178                 flow = flow_cast(flow_node);
1179                 if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1180                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1181                                            OVS_FLOW_CMD_NEW) < 0)
1182                         break;
1183
1184                 cb->args[0] = bucket;
1185                 cb->args[1] = obj;
1186         }
1187         return skb->len;
1188 }
1189
1190 static struct genl_ops dp_flow_genl_ops[] = {
1191         { .cmd = OVS_FLOW_CMD_NEW,
1192           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1193           .policy = flow_policy,
1194           .doit = ovs_flow_cmd_new_or_set
1195         },
1196         { .cmd = OVS_FLOW_CMD_DEL,
1197           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1198           .policy = flow_policy,
1199           .doit = ovs_flow_cmd_del
1200         },
1201         { .cmd = OVS_FLOW_CMD_GET,
1202           .flags = 0,               /* OK for unprivileged users. */
1203           .policy = flow_policy,
1204           .doit = ovs_flow_cmd_get,
1205           .dumpit = ovs_flow_cmd_dump
1206         },
1207         { .cmd = OVS_FLOW_CMD_SET,
1208           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1209           .policy = flow_policy,
1210           .doit = ovs_flow_cmd_new_or_set,
1211         },
1212 };
1213
1214 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1215 #ifdef HAVE_NLA_NUL_STRING
1216         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1217 #endif
1218         [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1219         [OVS_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1220 };
1221
1222 static struct genl_family dp_datapath_genl_family = {
1223         .id = GENL_ID_GENERATE,
1224         .hdrsize = sizeof(struct ovs_header),
1225         .name = OVS_DATAPATH_FAMILY,
1226         .version = 1,
1227         .maxattr = OVS_DP_ATTR_MAX
1228 };
1229
1230 static struct genl_multicast_group dp_datapath_multicast_group = {
1231         .name = OVS_DATAPATH_MCGROUP
1232 };
1233
1234 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1235                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1236 {
1237         struct ovs_header *ovs_header;
1238         struct nlattr *nla;
1239         int err;
1240
1241         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1242                                    flags, cmd);
1243         if (!ovs_header)
1244                 goto error;
1245
1246         ovs_header->dp_ifindex = dp->dp_ifindex;
1247
1248         rcu_read_lock();
1249         err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1250         rcu_read_unlock();
1251         if (err)
1252                 goto nla_put_failure;
1253
1254         nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1255         if (!nla)
1256                 goto nla_put_failure;
1257         get_dp_stats(dp, nla_data(nla));
1258
1259         NLA_PUT_U32(skb, OVS_DP_ATTR_IPV4_FRAGS,
1260                     dp->drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO);
1261
1262         if (dp->sflow_probability)
1263                 NLA_PUT_U32(skb, OVS_DP_ATTR_SAMPLING, dp->sflow_probability);
1264
1265         nla = nla_nest_start(skb, OVS_DP_ATTR_MCGROUPS);
1266         if (!nla)
1267                 goto nla_put_failure;
1268         NLA_PUT_U32(skb, OVS_PACKET_CMD_MISS, packet_mc_group(dp, OVS_PACKET_CMD_MISS));
1269         NLA_PUT_U32(skb, OVS_PACKET_CMD_ACTION, packet_mc_group(dp, OVS_PACKET_CMD_ACTION));
1270         NLA_PUT_U32(skb, OVS_PACKET_CMD_SAMPLE, packet_mc_group(dp, OVS_PACKET_CMD_SAMPLE));
1271         nla_nest_end(skb, nla);
1272
1273         return genlmsg_end(skb, ovs_header);
1274
1275 nla_put_failure:
1276         genlmsg_cancel(skb, ovs_header);
1277 error:
1278         return -EMSGSIZE;
1279 }
1280
1281 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1282                                              u32 seq, u8 cmd)
1283 {
1284         struct sk_buff *skb;
1285         int retval;
1286
1287         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1288         if (!skb)
1289                 return ERR_PTR(-ENOMEM);
1290
1291         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1292         if (retval < 0) {
1293                 kfree_skb(skb);
1294                 return ERR_PTR(retval);
1295         }
1296         return skb;
1297 }
1298
1299 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1300 {
1301         if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
1302                 u32 frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
1303
1304                 if (frags != OVS_DP_FRAG_ZERO && frags != OVS_DP_FRAG_DROP)
1305                         return -EINVAL;
1306         }
1307
1308         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1309 }
1310
1311 /* Called with genl_mutex and optionally with RTNL lock also. */
1312 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1313 {
1314         struct datapath *dp;
1315
1316         if (!a[OVS_DP_ATTR_NAME])
1317                 dp = get_dp(ovs_header->dp_ifindex);
1318         else {
1319                 struct vport *vport;
1320
1321                 rcu_read_lock();
1322                 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1323                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1324                 rcu_read_unlock();
1325         }
1326         return dp ? dp : ERR_PTR(-ENODEV);
1327 }
1328
1329 /* Called with genl_mutex. */
1330 static void change_datapath(struct datapath *dp, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1331 {
1332         if (a[OVS_DP_ATTR_IPV4_FRAGS])
1333                 dp->drop_frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]) == OVS_DP_FRAG_DROP;
1334         if (a[OVS_DP_ATTR_SAMPLING])
1335                 dp->sflow_probability = nla_get_u32(a[OVS_DP_ATTR_SAMPLING]);
1336 }
1337
1338 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1339 {
1340         struct nlattr **a = info->attrs;
1341         struct vport_parms parms;
1342         struct sk_buff *reply;
1343         struct datapath *dp;
1344         struct vport *vport;
1345         int err;
1346
1347         err = -EINVAL;
1348         if (!a[OVS_DP_ATTR_NAME])
1349                 goto err;
1350
1351         err = ovs_dp_cmd_validate(a);
1352         if (err)
1353                 goto err;
1354
1355         rtnl_lock();
1356         err = -ENODEV;
1357         if (!try_module_get(THIS_MODULE))
1358                 goto err_unlock_rtnl;
1359
1360         err = -ENOMEM;
1361         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1362         if (dp == NULL)
1363                 goto err_put_module;
1364         INIT_LIST_HEAD(&dp->port_list);
1365
1366         /* Initialize kobject for bridge.  This will be added as
1367          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1368         dp->ifobj.kset = NULL;
1369         kobject_init(&dp->ifobj, &dp_ktype);
1370
1371         /* Allocate table. */
1372         err = -ENOMEM;
1373         rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1374         if (!dp->table)
1375                 goto err_free_dp;
1376
1377         /* Set up our datapath device. */
1378         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1379         parms.type = OVS_VPORT_TYPE_INTERNAL;
1380         parms.options = NULL;
1381         parms.dp = dp;
1382         parms.port_no = OVSP_LOCAL;
1383         vport = new_vport(&parms);
1384         if (IS_ERR(vport)) {
1385                 err = PTR_ERR(vport);
1386                 if (err == -EBUSY)
1387                         err = -EEXIST;
1388
1389                 goto err_destroy_table;
1390         }
1391         dp->dp_ifindex = vport_get_ifindex(vport);
1392
1393         dp->drop_frags = 0;
1394         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1395         if (!dp->stats_percpu) {
1396                 err = -ENOMEM;
1397                 goto err_destroy_local_port;
1398         }
1399
1400         change_datapath(dp, a);
1401
1402         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1403         err = PTR_ERR(reply);
1404         if (IS_ERR(reply))
1405                 goto err_destroy_local_port;
1406
1407         list_add_tail(&dp->list_node, &dps);
1408         dp_sysfs_add_dp(dp);
1409
1410         rtnl_unlock();
1411
1412         genl_notify(reply, genl_info_net(info), info->snd_pid,
1413                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1414         return 0;
1415
1416 err_destroy_local_port:
1417         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1418 err_destroy_table:
1419         tbl_destroy(get_table_protected(dp), NULL);
1420 err_free_dp:
1421         kfree(dp);
1422 err_put_module:
1423         module_put(THIS_MODULE);
1424 err_unlock_rtnl:
1425         rtnl_unlock();
1426 err:
1427         return err;
1428 }
1429
1430 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1431 {
1432         struct vport *vport, *next_vport;
1433         struct sk_buff *reply;
1434         struct datapath *dp;
1435         int err;
1436
1437         err = ovs_dp_cmd_validate(info->attrs);
1438         if (err)
1439                 goto exit;
1440
1441         rtnl_lock();
1442         dp = lookup_datapath(info->userhdr, info->attrs);
1443         err = PTR_ERR(dp);
1444         if (IS_ERR(dp))
1445                 goto exit_unlock;
1446
1447         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
1448         err = PTR_ERR(reply);
1449         if (IS_ERR(reply))
1450                 goto exit_unlock;
1451
1452         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1453                 if (vport->port_no != OVSP_LOCAL)
1454                         dp_detach_port(vport);
1455
1456         dp_sysfs_del_dp(dp);
1457         list_del(&dp->list_node);
1458         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1459
1460         /* rtnl_unlock() will wait until all the references to devices that
1461          * are pending unregistration have been dropped.  We do it here to
1462          * ensure that any internal devices (which contain DP pointers) are
1463          * fully destroyed before freeing the datapath.
1464          */
1465         rtnl_unlock();
1466
1467         call_rcu(&dp->rcu, destroy_dp_rcu);
1468         module_put(THIS_MODULE);
1469
1470         genl_notify(reply, genl_info_net(info), info->snd_pid,
1471                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1472
1473         return 0;
1474
1475 exit_unlock:
1476         rtnl_unlock();
1477 exit:
1478         return err;
1479 }
1480
1481 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1482 {
1483         struct sk_buff *reply;
1484         struct datapath *dp;
1485         int err;
1486
1487         err = ovs_dp_cmd_validate(info->attrs);
1488         if (err)
1489                 return err;
1490
1491         dp = lookup_datapath(info->userhdr, info->attrs);
1492         if (IS_ERR(dp))
1493                 return PTR_ERR(dp);
1494
1495         change_datapath(dp, info->attrs);
1496
1497         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1498         if (IS_ERR(reply)) {
1499                 err = PTR_ERR(reply);
1500                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1501                                 dp_datapath_multicast_group.id, err);
1502                 return 0;
1503         }
1504
1505         genl_notify(reply, genl_info_net(info), info->snd_pid,
1506                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1507         return 0;
1508 }
1509
1510 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1511 {
1512         struct sk_buff *reply;
1513         struct datapath *dp;
1514         int err;
1515
1516         err = ovs_dp_cmd_validate(info->attrs);
1517         if (err)
1518                 return err;
1519
1520         dp = lookup_datapath(info->userhdr, info->attrs);
1521         if (IS_ERR(dp))
1522                 return PTR_ERR(dp);
1523
1524         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1525         if (IS_ERR(reply))
1526                 return PTR_ERR(reply);
1527
1528         return genlmsg_reply(reply, info);
1529 }
1530
1531 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1532 {
1533         struct datapath *dp;
1534         int skip = cb->args[0];
1535         int i = 0;
1536
1537         list_for_each_entry (dp, &dps, list_node) {
1538                 if (i < skip)
1539                         continue;
1540                 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1541                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1542                                          OVS_DP_CMD_NEW) < 0)
1543                         break;
1544                 i++;
1545         }
1546
1547         cb->args[0] = i;
1548
1549         return skb->len;
1550 }
1551
1552 static struct genl_ops dp_datapath_genl_ops[] = {
1553         { .cmd = OVS_DP_CMD_NEW,
1554           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1555           .policy = datapath_policy,
1556           .doit = ovs_dp_cmd_new
1557         },
1558         { .cmd = OVS_DP_CMD_DEL,
1559           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1560           .policy = datapath_policy,
1561           .doit = ovs_dp_cmd_del
1562         },
1563         { .cmd = OVS_DP_CMD_GET,
1564           .flags = 0,               /* OK for unprivileged users. */
1565           .policy = datapath_policy,
1566           .doit = ovs_dp_cmd_get,
1567           .dumpit = ovs_dp_cmd_dump
1568         },
1569         { .cmd = OVS_DP_CMD_SET,
1570           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1571           .policy = datapath_policy,
1572           .doit = ovs_dp_cmd_set,
1573         },
1574 };
1575
1576 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1577 #ifdef HAVE_NLA_NUL_STRING
1578         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1579         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1580         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1581         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1582         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1583 #else
1584         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1585         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1586 #endif
1587         [OVS_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1588         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1589 };
1590
1591 static struct genl_family dp_vport_genl_family = {
1592         .id = GENL_ID_GENERATE,
1593         .hdrsize = sizeof(struct ovs_header),
1594         .name = OVS_VPORT_FAMILY,
1595         .version = 1,
1596         .maxattr = OVS_VPORT_ATTR_MAX
1597 };
1598
1599 struct genl_multicast_group dp_vport_multicast_group = {
1600         .name = OVS_VPORT_MCGROUP
1601 };
1602
1603 /* Called with RTNL lock or RCU read lock. */
1604 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1605                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1606 {
1607         struct ovs_header *ovs_header;
1608         struct nlattr *nla;
1609         int ifindex;
1610         int mtu;
1611         int err;
1612
1613         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1614                                  flags, cmd);
1615         if (!ovs_header)
1616                 return -EMSGSIZE;
1617
1618         ovs_header->dp_ifindex = vport->dp->dp_ifindex;
1619
1620         NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1621         NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1622         NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1623
1624         nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1625         if (!nla)
1626                 goto nla_put_failure;
1627         if (vport_get_stats(vport, nla_data(nla)))
1628                 __skb_trim(skb, skb->len - nla->nla_len);
1629
1630         NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1631
1632         mtu = vport_get_mtu(vport);
1633         if (mtu)
1634                 NLA_PUT_U32(skb, OVS_VPORT_ATTR_MTU, mtu);
1635
1636         err = vport_get_options(vport, skb);
1637         if (err == -EMSGSIZE)
1638                 goto error;
1639
1640         ifindex = vport_get_ifindex(vport);
1641         if (ifindex > 0)
1642                 NLA_PUT_U32(skb, OVS_VPORT_ATTR_IFINDEX, ifindex);
1643
1644         return genlmsg_end(skb, ovs_header);
1645
1646 nla_put_failure:
1647         err = -EMSGSIZE;
1648 error:
1649         genlmsg_cancel(skb, ovs_header);
1650         return err;
1651 }
1652
1653 /* Called with RTNL lock or RCU read lock. */
1654 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1655                                          u32 seq, u8 cmd)
1656 {
1657         struct sk_buff *skb;
1658         int retval;
1659
1660         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1661         if (!skb)
1662                 return ERR_PTR(-ENOMEM);
1663
1664         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1665         if (retval < 0) {
1666                 kfree_skb(skb);
1667                 return ERR_PTR(retval);
1668         }
1669         return skb;
1670 }
1671
1672 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1673 {
1674         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1675 }
1676
1677 /* Called with RTNL lock or RCU read lock. */
1678 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1679                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1680 {
1681         struct datapath *dp;
1682         struct vport *vport;
1683
1684         if (a[OVS_VPORT_ATTR_NAME]) {
1685                 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1686                 if (!vport)
1687                         return ERR_PTR(-ENODEV);
1688                 return vport;
1689         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1690                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1691
1692                 if (port_no >= DP_MAX_PORTS)
1693                         return ERR_PTR(-EFBIG);
1694
1695                 dp = get_dp(ovs_header->dp_ifindex);
1696                 if (!dp)
1697                         return ERR_PTR(-ENODEV);
1698
1699                 vport = get_vport_protected(dp, port_no);
1700                 if (!vport)
1701                         return ERR_PTR(-ENOENT);
1702                 return vport;
1703         } else
1704                 return ERR_PTR(-EINVAL);
1705 }
1706
1707 /* Called with RTNL lock. */
1708 static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1709 {
1710         int err = 0;
1711         if (a[OVS_VPORT_ATTR_STATS])
1712                 err = vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1713         if (!err && a[OVS_VPORT_ATTR_ADDRESS])
1714                 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1715         if (!err && a[OVS_VPORT_ATTR_MTU])
1716                 err = vport_set_mtu(vport, nla_get_u32(a[OVS_VPORT_ATTR_MTU]));
1717         return err;
1718 }
1719
1720 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1721 {
1722         struct nlattr **a = info->attrs;
1723         struct ovs_header *ovs_header = info->userhdr;
1724         struct vport_parms parms;
1725         struct sk_buff *reply;
1726         struct vport *vport;
1727         struct datapath *dp;
1728         u32 port_no;
1729         int err;
1730
1731         err = -EINVAL;
1732         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE])
1733                 goto exit;
1734
1735         err = ovs_vport_cmd_validate(a);
1736         if (err)
1737                 goto exit;
1738
1739         rtnl_lock();
1740         dp = get_dp(ovs_header->dp_ifindex);
1741         err = -ENODEV;
1742         if (!dp)
1743                 goto exit_unlock;
1744
1745         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1746                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1747
1748                 err = -EFBIG;
1749                 if (port_no >= DP_MAX_PORTS)
1750                         goto exit_unlock;
1751
1752                 vport = get_vport_protected(dp, port_no);
1753                 err = -EBUSY;
1754                 if (vport)
1755                         goto exit_unlock;
1756         } else {
1757                 for (port_no = 1; ; port_no++) {
1758                         if (port_no >= DP_MAX_PORTS) {
1759                                 err = -EFBIG;
1760                                 goto exit_unlock;
1761                         }
1762                         vport = get_vport_protected(dp, port_no);
1763                         if (!vport)
1764                                 break;
1765                 }
1766         }
1767
1768         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1769         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1770         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1771         parms.dp = dp;
1772         parms.port_no = port_no;
1773
1774         vport = new_vport(&parms);
1775         err = PTR_ERR(vport);
1776         if (IS_ERR(vport))
1777                 goto exit_unlock;
1778
1779         set_internal_devs_mtu(dp);
1780         dp_sysfs_add_if(vport);
1781
1782         err = change_vport(vport, a);
1783         if (!err) {
1784                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1785                                                  info->snd_seq, OVS_VPORT_CMD_NEW);
1786                 if (IS_ERR(reply))
1787                         err = PTR_ERR(reply);
1788         }
1789         if (err) {
1790                 dp_detach_port(vport);
1791                 goto exit_unlock;
1792         }
1793         genl_notify(reply, genl_info_net(info), info->snd_pid,
1794                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1795
1796
1797 exit_unlock:
1798         rtnl_unlock();
1799 exit:
1800         return err;
1801 }
1802
1803 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1804 {
1805         struct nlattr **a = info->attrs;
1806         struct sk_buff *reply;
1807         struct vport *vport;
1808         int err;
1809
1810         err = ovs_vport_cmd_validate(a);
1811         if (err)
1812                 goto exit;
1813
1814         rtnl_lock();
1815         vport = lookup_vport(info->userhdr, a);
1816         err = PTR_ERR(vport);
1817         if (IS_ERR(vport))
1818                 goto exit_unlock;
1819
1820         err = 0;
1821         if (a[OVS_VPORT_ATTR_OPTIONS])
1822                 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1823         if (!err)
1824                 err = change_vport(vport, a);
1825
1826         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1827                                          OVS_VPORT_CMD_NEW);
1828         if (IS_ERR(reply)) {
1829                 err = PTR_ERR(reply);
1830                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1831                                 dp_vport_multicast_group.id, err);
1832                 return 0;
1833         }
1834
1835         genl_notify(reply, genl_info_net(info), info->snd_pid,
1836                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1837
1838 exit_unlock:
1839         rtnl_unlock();
1840 exit:
1841         return err;
1842 }
1843
1844 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1845 {
1846         struct nlattr **a = info->attrs;
1847         struct sk_buff *reply;
1848         struct vport *vport;
1849         int err;
1850
1851         err = ovs_vport_cmd_validate(a);
1852         if (err)
1853                 goto exit;
1854
1855         rtnl_lock();
1856         vport = lookup_vport(info->userhdr, a);
1857         err = PTR_ERR(vport);
1858         if (IS_ERR(vport))
1859                 goto exit_unlock;
1860
1861         if (vport->port_no == OVSP_LOCAL) {
1862                 err = -EINVAL;
1863                 goto exit_unlock;
1864         }
1865
1866         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1867                                          OVS_VPORT_CMD_DEL);
1868         err = PTR_ERR(reply);
1869         if (IS_ERR(reply))
1870                 goto exit_unlock;
1871
1872         err = dp_detach_port(vport);
1873
1874         genl_notify(reply, genl_info_net(info), info->snd_pid,
1875                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1876
1877 exit_unlock:
1878         rtnl_unlock();
1879 exit:
1880         return err;
1881 }
1882
1883 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1884 {
1885         struct nlattr **a = info->attrs;
1886         struct ovs_header *ovs_header = info->userhdr;
1887         struct sk_buff *reply;
1888         struct vport *vport;
1889         int err;
1890
1891         err = ovs_vport_cmd_validate(a);
1892         if (err)
1893                 goto exit;
1894
1895         rcu_read_lock();
1896         vport = lookup_vport(ovs_header, a);
1897         err = PTR_ERR(vport);
1898         if (IS_ERR(vport))
1899                 goto exit_unlock;
1900
1901         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1902                                          OVS_VPORT_CMD_NEW);
1903         err = PTR_ERR(reply);
1904         if (IS_ERR(reply))
1905                 goto exit_unlock;
1906
1907         rcu_read_unlock();
1908
1909         return genlmsg_reply(reply, info);
1910
1911 exit_unlock:
1912         rcu_read_unlock();
1913 exit:
1914         return err;
1915 }
1916
1917 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1918 {
1919         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1920         struct datapath *dp;
1921         u32 port_no;
1922         int retval;
1923
1924         dp = get_dp(ovs_header->dp_ifindex);
1925         if (!dp)
1926                 return -ENODEV;
1927
1928         rcu_read_lock();
1929         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1930                 struct vport *vport;
1931
1932                 vport = get_vport_protected(dp, port_no);
1933                 if (!vport)
1934                         continue;
1935
1936                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1937                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1938                                             OVS_VPORT_CMD_NEW) < 0)
1939                         break;
1940         }
1941         rcu_read_unlock();
1942
1943         cb->args[0] = port_no;
1944         retval = skb->len;
1945
1946         return retval;
1947 }
1948
1949 static struct genl_ops dp_vport_genl_ops[] = {
1950         { .cmd = OVS_VPORT_CMD_NEW,
1951           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1952           .policy = vport_policy,
1953           .doit = ovs_vport_cmd_new
1954         },
1955         { .cmd = OVS_VPORT_CMD_DEL,
1956           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1957           .policy = vport_policy,
1958           .doit = ovs_vport_cmd_del
1959         },
1960         { .cmd = OVS_VPORT_CMD_GET,
1961           .flags = 0,               /* OK for unprivileged users. */
1962           .policy = vport_policy,
1963           .doit = ovs_vport_cmd_get,
1964           .dumpit = ovs_vport_cmd_dump
1965         },
1966         { .cmd = OVS_VPORT_CMD_SET,
1967           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1968           .policy = vport_policy,
1969           .doit = ovs_vport_cmd_set,
1970         },
1971 };
1972
1973 struct genl_family_and_ops {
1974         struct genl_family *family;
1975         struct genl_ops *ops;
1976         int n_ops;
1977         struct genl_multicast_group *group;
1978 };
1979
1980 static const struct genl_family_and_ops dp_genl_families[] = {
1981         { &dp_datapath_genl_family,
1982           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1983           &dp_datapath_multicast_group },
1984         { &dp_vport_genl_family,
1985           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1986           &dp_vport_multicast_group },
1987         { &dp_flow_genl_family,
1988           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1989           &dp_flow_multicast_group },
1990         { &dp_packet_genl_family,
1991           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1992           NULL },
1993 };
1994
1995 static void dp_unregister_genl(int n_families)
1996 {
1997         int i;
1998
1999         for (i = 0; i < n_families; i++)
2000                 genl_unregister_family(dp_genl_families[i].family);
2001 }
2002
2003 static int dp_register_genl(void)
2004 {
2005         int n_registered;
2006         int err;
2007         int i;
2008
2009         n_registered = 0;
2010         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2011                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2012
2013                 err = genl_register_family_with_ops(f->family, f->ops,
2014                                                     f->n_ops);
2015                 if (err)
2016                         goto error;
2017                 n_registered++;
2018
2019                 if (f->group) {
2020                         err = genl_register_mc_group(f->family, f->group);
2021                         if (err)
2022                                 goto error;
2023                 }
2024         }
2025
2026         err = packet_register_mc_groups();
2027         if (err)
2028                 goto error;
2029         return 0;
2030
2031 error:
2032         dp_unregister_genl(n_registered);
2033         return err;
2034 }
2035
2036 static int __init dp_init(void)
2037 {
2038         struct sk_buff *dummy_skb;
2039         int err;
2040
2041         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2042
2043         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2044
2045         err = flow_init();
2046         if (err)
2047                 goto error;
2048
2049         err = vport_init();
2050         if (err)
2051                 goto error_flow_exit;
2052
2053         err = register_netdevice_notifier(&dp_device_notifier);
2054         if (err)
2055                 goto error_vport_exit;
2056
2057         err = dp_register_genl();
2058         if (err < 0)
2059                 goto error_unreg_notifier;
2060
2061         return 0;
2062
2063 error_unreg_notifier:
2064         unregister_netdevice_notifier(&dp_device_notifier);
2065 error_vport_exit:
2066         vport_exit();
2067 error_flow_exit:
2068         flow_exit();
2069 error:
2070         return err;
2071 }
2072
2073 static void dp_cleanup(void)
2074 {
2075         rcu_barrier();
2076         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2077         unregister_netdevice_notifier(&dp_device_notifier);
2078         vport_exit();
2079         flow_exit();
2080 }
2081
2082 module_init(dp_init);
2083 module_exit(dp_cleanup);
2084
2085 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2086 MODULE_LICENSE("GPL");